A Blood Test That Sorts Cancer Referrals, Tested but Not Yet Used

experiment · Mayo Clinic proceedings. Digital health · la publicación, 10 jun 2026 · gratis

Una prueba de sangre ordenó a los pacientes por riesgo de cáncer, pero nadie se curó por ella todavía

El estudio se hizo en una sola región de Inglaterra y la prueba nunca llegó a decidir el tratamiento de nadie.

Versión breve · la versión detallada sigue, unos 6 min

Pregunte al chat de weeklyAI

Pregúnteme por este estudio: a quiénes se estudió, qué encontró y qué no dice.

Las conversaciones se guardan mientras exista weeklyAI, para mejorar la publicación. Se responde en el idioma en que usted escribe.

El estudio, de un vistazo
Quiénes: Pacientes derivados por sospecha urgente de cáncer
Cuántos: 16,481 pacientes
Dónde: West Yorkshire y Harrogate, Inglaterra
Cuándo: De diciembre de 2020 a julio de 2025
Tipo de estudio: análisis de lo que hicieron las personas
Quién lo hizo: PinPoint Data Science, universidades y hospitales del NHS
El límite que importa: La prueba no se usó para decidir el tratamiento de nadie.

La prueba se calculaba y se guardaba en un laboratorio, pero los médicos no la veían y nadie fue tratado con base en ella.
Lectura de weeklyAI

Así podría verse · ilustración generada por weeklyAI.watch, no es una fotografía

Imagine que su médico lo manda de urgencia a descartar un cáncer. Empieza entonces una ruta con muchos exámenes y muchas esperas. En Inglaterra, esa ruta se llama derivación urgente por sospecha de cáncer, y cada año entran más de tres millones de personas.

Entre 2020 y 2025, un equipo de investigación probó nueve pruebas de sangre distintas en 16,481 pacientes que entraban por esa vía. Cada prueba correspondía a una ruta de sospecha diferente. De todos ellos, 13,255 quedaron en el análisis principal. Trabajaron con cinco hospitales y 170 consultorios de una región del norte de Inglaterra, West Yorkshire y Harrogate.

La prueba sí medía marcadores tumorales en la sangre, junto con otros análisis comunes. Usaba la edad, el sexo y los resultados de análisis de sangre comunes, los mismos que ya se piden, y con ellos calculaba qué tan probable era que esa persona tuviera cáncer. Un programa de computadora hacía el cálculo.

En cinco rutas, la prueba ordenó bien a los pacientes. Si se tomaba al 10% con el puntaje más alto, ese grupo concentraba tantos cánceres que los médicos necesitaban hacer entre 2.6 y 6.1 veces menos exámenes para encontrar un caso.

Los resultados fueron mejores en la ruta del aparato digestivo alto. Cuando la prueba eligió al 10% de mayor riesgo, encontró al 61% de los cánceres de esa ruta. En la ruta ginecológica encontró al 41% de los cánceres. En la de pulmón encontró al 26% de los cánceres.

Ninguna de esas cifras cambió la vida de un paciente durante el estudio. La prueba se calculaba y se guardaba en un laboratorio, pero los médicos no la veían y nadie fue tratado con base en ella. Por eso lo que muestra es una posibilidad, no un beneficio ya comprobado.

Tampoco puede decirse que funcione igual en todos los grupos: faltaban datos de origen étnico en la mayoría de los pacientes. Las pruebas de mama y piel no mejoraron mucho lo que ya se sabía por la edad. La de sangre todavía no está clara. La de urología no mejoró de forma clara lo que ya lograba el análisis de PSA.

Si algún día llega a su país, una prueba así podría servir para que su médico decida quién necesita exámenes más rápido y quién puede seguir en observación tranquila. No reemplaza los exámenes que su médico le indique.

Vale la pena estar atento a una noticia concreta: que la prueba se esté usando en la atención de rutina y que alguien esté midiendo si los pacientes se diagnostican antes.

Qué significa para usted

Lo que esto significa para usted es sencillo: por ahora no hay nada que pedir en su clínica. La prueba mostró que puede ordenar bien a los pacientes en cinco rutas de sospecha, pero nadie fue tratado con base en ella, así que no mejoró a ninguna persona. Si algún día la noticia es que ya se usa en la atención de rutina, pregunte a su médico si eso cambió algo.

Neal, M., Dean, M., Duffy, S. et al. (2026). Real-World Validation of PinPoint Blood Tests in the NHS: Multivariable Machine Learning to Predict Cancer Risk in Primary Care Urgent Referrals. Mayo Clinic Proceedings: Digital Health. https://doi.org/10.1016/j.mcpdig.2026.100382

Quién pagó: PinPoint Data Science financió las contribuciones de varios autores, y aspectos del trabajo fueron apoyados por un premio de SBRI Healthcare y la West Yorkshire and Harrogate Cancer Alliance; el artículo no indica si los financiadores tuvieron alguna influencia en el estudio.

No tome esto como consejo médico profesional.

Los hallazgos de otros estudios que aquí se mencionan los conocemos por este documento, que fue el que leímos; no abrimos cada uno de esos estudios.

Versión detalladaLos pasajes copiados del artículo, las ilustraciones y cada fuente con cuánto leímos de ella · 1226 palabras · unos 6 minLeerla →Cerrar

Un análisis de sangre con datos que ya existen podría ordenar las urgencias oncológicas. Nadie lo usó todavía para tratar a nadie.

Un gran estudio en Inglaterra probó nueve pruebas de riesgo de cáncer en personas con sospecha urgente. Cinco mostraron resultados prometedores. La diferencia con la atención real todavía no se midió.

Cuando un médico de cabecera sospecha cáncer, envía al paciente por una vía de urgencia. En Inglaterra esas vías reciben más de 3 millones de personas al año, la cifra crece un 10% anual desde hace quince años, y solo el 6% de los derivados termina con un diagnóstico de cáncer¹. Es decir: muchísimas personas pasan por pruebas invasivas para encontrar pocos casos.

La idea que se evaluó es distinta de una prueba que "detecta cáncer". Es un conjunto de nueve pruebas de sangre con marca de conformidad del Reino Unido, pensadas para estimar el riesgo de cáncer en personas que ya tienen síntomas². No buscan una sustancia nueva: usan la edad, el sexo y análisis que ya se piden de rutina —hemograma, función hepática, urea y electrolitos, perfil óseo, marcadores de inflamación y algunos marcadores tumorales— y un modelo de aprendizaje automático los combina en un puntaje de riesgo³.

Así funciona, paso a paso: al paciente se le toma sangre al inicio de su derivación; la muestra se procesa en un laboratorio de rutina; los valores entran al algoritmo alojado en infraestructura informática del sistema de salud; el resultado vuelve como una probabilidad. En este estudio los médicos no vieron ese resultado y el recorrido del paciente no cambió⁴.

El trabajo reunió a 16,481 pacientes derivados por sospecha urgente en cinco hospitales y 170 consultorios de medicina general de West Yorkshire y Harrogate, entre diciembre de 2020 y julio de 2025⁵. Fue un estudio observacional, prospectivo, de servicio real: las tomas de sangre y la logística se hicieron como se harían en la práctica cotidiana⁶.

Cinco de las nueve pruebas mostraron desempeño con posible utilidad clínica. Las cifras de acierto (el área bajo la curva, donde 0,5 es azar y 1 es perfecto) fueron: vía gastrointestinal alta 0,86; ginecológica 0,81; pulmón 0,79; cabeza y cuello 0,73; gastrointestinal baja 0,72⁷. Si se priorizara al 10% de mayor riesgo, el número de personas que hay que investigar para encontrar un cáncer bajaría entre 2,6 y 6,1 veces⁷. Cuatro de esas cinco pruebas dejaron pasar como "de bajo riesgo" a un 20% de los pacientes sin enfermedad con una certeza superior al 99%⁸.

Los límites importan tanto como el resultado. Los resultados del algoritmo no se usaron clínicamente: se calcularon y se guardaron para agregarlos, nada más⁴. Unas 3,197 personas, el 19.4% de las inscritas, quedaron fuera del análisis principal por diversas razones⁹. Los datos de etnia faltaban en una proporción grande de pacientes en casi todas las vías, así que no se puede concluir nada sobre el desempeño entre grupos étnicos. No todas las vías llegaron a 100 cánceres confirmados. Los criterios de derivación cambiaron durante el estudio. Y el trabajo empezó en pandemia, lo que alteró cifras y características de quienes llegaron.

Sobre quién pagó: varios autores trabajan en PinPoint Data Science y son accionistas o tenedores de opciones; la empresa financió las contribuciones de varios de ellos; la Universidad de Leeds y un hospital tienen un acuerdo de regalías con la empresa, y uno de los autores figura como inventor en ese acuerdo¹⁰.

Para dimensionar la apuesta: la vía gastrointestinal alta, la de mejor desempeño, recibe unas 250,000 derivaciones al año con un rendimiento de alrededor del 3% y hoy no tiene herramienta de triaje¹¹. La de cabeza y cuello recibe unas 285,000 al año, tampoco tiene triaje rutinario, y solo un tercio de sus cánceres se diagnostica en etapa temprana¹². La prueba ginecológica se validó sobre todo contra cáncer de endometrio y podría reducir la demanda de histeroscopias, de las que se hacen unas 100,000 al año en Inglaterra con un rendimiento de cáncer del 5% al 10%¹³.

En la misma familia de ideas, otros grupos trabajan con herramientas parecidas. Un equipo desarrolló una plataforma que combina aprendizaje automático con una técnica de laboratorio muy establecida para detectar una proteína anómala en la sangre, señal temprana de un cáncer de la médula; tras entrenarla y validarla con más de cinco mil muestras, la desplegó en algo más de doce mil pacientes adultos, según el resumen de ese trabajo, que pudimos leer pero cuya versión completa está detrás de una suscripción¹⁴¹⁵¹⁶. Otro equipo, en China, propuso un sistema que lee imágenes de biopsias al microscopio para reconocer un subtipo agresivo de cáncer de cuello uterino que suele pasar desapercibido, y lo probó después en miles de casos reales; también leímos solo el resumen¹⁷¹⁸¹⁹. Y un tercer grupo evaluó un sistema digital que acompaña a pacientes en quimioterapia entre una consulta y la siguiente, con seguimiento y alertas; los propios autores piden estudios aleatorizados para confirmar lo que observaron²⁰²¹²².

La promesa concreta que el artículo deja sobre la mesa: pacientes de mayor riesgo podrían avanzar más rápido por la vía, y los de menor riesgo podrían evitar pruebas invasivas innecesarias; y el software podría desplegarse sin equipos nuevos²³. Nada de eso se midió aquí. Es lo que el estudio vuelve posible explorar, no lo que demostró.

Así lo leemos nosotros. Lo notable de este trabajo no es el puntaje: es de dónde sale. La prueba no pide nada que no se pida ya. Eso la vuelve, en principio, transportable a lugares donde falta especialista pero no falta laboratorio básico. Lo que no podemos saber, y conviene decirlo claro, es si funcionará igual fuera de esa región inglesa y de ese sistema de salud, porque nadie lo probó ahí. También notamos una ausencia que se repite en este tipo de herramientas: los datos de etnia faltaban en la mayoría de los pacientes, así que no hay forma de saber si el puntaje se comporta igual para todos. Cuando un sistema automático decide con datos incompletos, tiende a reproducir las desigualdades que ya existían, porque nadie revisó a quién estaba dejando afuera. No hay aquí ningún dato que confirme ni descarte eso; simplemente no se puede comprobar. Si quisiéramos saber que nos equivocamos, bastaría con un estudio que trajera datos completos de todos los grupos y resultados parecidos entre ellos.

Y una segunda cosa que vale para usted, esté donde esté. Un puntaje que ordena pacientes no dice quién responde si se equivoca, ni cómo se pide una segunda opinión fuera del número. El estudio no incluye ningún mecanismo para que un paciente o su médico cuestionen el resultado, ni explica quién asume la responsabilidad cuando falla. Eso no es un defecto de este trabajo en particular: es lo que suele faltar en esta familia de herramientas. Si algún día le ofrecen algo así, pida que le expliquen en palabras simples qué significa su resultado y qué pasa si no está de acuerdo. Guarde el nombre de la persona o el servicio que puede revisar su caso. Un número no debería decidir solo.

Para terminar, lo que este trabajo deja en sus manos es una pregunta útil para llevar a su próxima consulta: si le derivan por sospecha de cáncer, pregunte si su caso puede ordenarse con los análisis que ya le mandaron, o si hay que viajar o pagar algo aparte, y por qué.

De dónde sale cada dato de contexto, y cuánto leímos de cada documento

Neal, M., Dean, M., Duffy, S. et al. (2026). Real-World Validation of PinPoint Blood Tests in the NHS: Multivariable Machine Learning to Predict Cancer Risk in Primary Care Urgent Referrals. Mayo Clinic Proceedings: Digital Health. https://doi.org/10.1016/j.mcpdig.2026.100382 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Urgent referrals from primary care are a major route for cancer diagnosis in the National Health service (NHS), with more than 3 million patients referred annually in England alone. The annual referral rate has increased 10% year-on-year for the last 15 years, 1 and the urgent suspected cancer (USC) pathways have an average conversion rate of only 6%, meaning that improved methods of triage are urgently needed."
Neal, M., Dean, M., Duffy, S. et al. (2026). Real-World Validation of PinPoint Blood Tests in the NHS: Multivariable Machine Learning to Predict Cancer Risk in Primary Care Urgent Referrals. Mayo Clinic Proceedings: Digital Health. https://doi.org/10.1016/j.mcpdig.2026.100382 - el artículo del que trata esta nota — el artículo completo — el pasaje: "The PinPoint Tests are a set of United Kingdom Conformity Assessed-marked multi-cancer early detection blood tests for predicting the cancer risk of symptomatic patients."
Neal, M., Dean, M., Duffy, S. et al. (2026). Real-World Validation of PinPoint Blood Tests in the NHS: Multivariable Machine Learning to Predict Cancer Risk in Primary Care Urgent Referrals. Mayo Clinic Proceedings: Digital Health. https://doi.org/10.1016/j.mcpdig.2026.100382 - el artículo del que trata esta nota — el artículo completo — el pasaje: "The predictors are age, sex, and a panel of standard blood analytes, including full blood count, liver function tests, urea and electrolytes, bone profile, inflammatory markers, and a set of tumor markers."
Neal, M., Dean, M., Duffy, S. et al. (2026). Real-World Validation of PinPoint Blood Tests in the NHS: Multivariable Machine Learning to Predict Cancer Risk in Primary Care Urgent Referrals. Mayo Clinic Proceedings: Digital Health. https://doi.org/10.1016/j.mcpdig.2026.100382 - el artículo del que trata esta nota — el artículo completo — el pasaje: "The evaluation was observational. PinPoint Test results were calculated and returned to the regional hub laboratory for aggregation but were not used clinically."
Neal, M., Dean, M., Duffy, S. et al. (2026). Real-World Validation of PinPoint Blood Tests in the NHS: Multivariable Machine Learning to Predict Cancer Risk in Primary Care Urgent Referrals. Mayo Clinic Proceedings: Digital Health. https://doi.org/10.1016/j.mcpdig.2026.100382 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Total of 16,481 patients with urgent suspected cancer referrals were enrolled across 5 secondary care Trusts and 170 General Practitioner surgeries."
Neal, M., Dean, M., Duffy, S. et al. (2026). Real-World Validation of PinPoint Blood Tests in the NHS: Multivariable Machine Learning to Predict Cancer Risk in Primary Care Urgent Referrals. Mayo Clinic Proceedings: Digital Health. https://doi.org/10.1016/j.mcpdig.2026.100382 - el artículo del que trata esta nota — el artículo completo — el pasaje: "This work comprises a large-scale, prospective, observational, real-world NHS service evaluation of nine United Kingdom Conformity Assessed-marked Software as a Medical Device multi-cancer early detection blood tests."
Neal, M., Dean, M., Duffy, S. et al. (2026). Real-World Validation of PinPoint Blood Tests in the NHS: Multivariable Machine Learning to Predict Cancer Risk in Primary Care Urgent Referrals. Mayo Clinic Proceedings: Digital Health. https://doi.org/10.1016/j.mcpdig.2026.100382 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Five tests have performance indicating potential clinical utility. Receiver operating characteristic area-under-curve scores (95% CI) for these were: Upper gastrointestinal=0.86 (0.81-0.90), Gynecological=0.81 (0.77-0.85), Lung=0.79 (0.74-0.84), Head & Neck=0.73 (0.68-0.78), and Lower gastrointestinal=0.72 (0.67-0.78), Prioritization of the 10% of highest-risk patients would reduce the number needed to investigate to detect one cancer by a factor of 2.6-6.1."
Neal, M., Dean, M., Duffy, S. et al. (2026). Real-World Validation of PinPoint Blood Tests in the NHS: Multivariable Machine Learning to Predict Cancer Risk in Primary Care Urgent Referrals. Mayo Clinic Proceedings: Digital Health. https://doi.org/10.1016/j.mcpdig.2026.100382 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Four of these five tests achieved an negative predictive value >0.99 when used to rule-out 20% of the lowest-risk patients."
Neal, M., Dean, M., Duffy, S. et al. (2026). Real-World Validation of PinPoint Blood Tests in the NHS: Multivariable Machine Learning to Predict Cancer Risk in Primary Care Urgent Referrals. Mayo Clinic Proceedings: Digital Health. https://doi.org/10.1016/j.mcpdig.2026.100382 - el artículo del que trata esta nota — el artículo completo — el pasaje: "About 3,197 (19.4%) patients were excluded from the analysis for various reasons."
Neal, M., Dean, M., Duffy, S. et al. (2026). Real-World Validation of PinPoint Blood Tests in the NHS: Multivariable Machine Learning to Predict Cancer Risk in Primary Care Urgent Referrals. Mayo Clinic Proceedings: Digital Health. https://doi.org/10.1016/j.mcpdig.2026.100382 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Drs Savage, Lloyd, Neal, Skinner, Sansom, Tully, Ferguson, and Duffy are employed by, and are shareholders or option holders in, PinPoint Data Science. Both the University of Leeds and Leeds Teaching Hospitals Trust have a royalty agreement with PinPoint Data Science; Prof Richard Neal is a named inventor in this royalty agreement."
Neal, M., Dean, M., Duffy, S. et al. (2026). Real-World Validation of PinPoint Blood Tests in the NHS: Multivariable Machine Learning to Predict Cancer Risk in Primary Care Urgent Referrals. Mayo Clinic Proceedings: Digital Health. https://doi.org/10.1016/j.mcpdig.2026.100382 - el artículo del que trata esta nota — el artículo completo — el pasaje: "The Upper GI test, which has the strongest performance, addresses a high-volume (250,000 referrals/year), low-yield (∼3%) pathway with no current triage tool."
Neal, M., Dean, M., Duffy, S. et al. (2026). Real-World Validation of PinPoint Blood Tests in the NHS: Multivariable Machine Learning to Predict Cancer Risk in Primary Care Urgent Referrals. Mayo Clinic Proceedings: Digital Health. https://doi.org/10.1016/j.mcpdig.2026.100382 - el artículo del que trata esta nota — el artículo completo — el pasaje: "The Head & Neck test addresses another high-volume (285,000 referrals/year) pathway with no routine triage tool, where only one third of cancers are currently diagnosed early and only half start treatment within the NHS 62-day target."
Neal, M., Dean, M., Duffy, S. et al. (2026). Real-World Validation of PinPoint Blood Tests in the NHS: Multivariable Machine Learning to Predict Cancer Risk in Primary Care Urgent Referrals. Mayo Clinic Proceedings: Digital Health. https://doi.org/10.1016/j.mcpdig.2026.100382 - el artículo del que trata esta nota — el artículo completo — el pasaje: "The gynecological test (validated primarily against endometrial cancer, 61% of cohort cancers) could reduce demand for hysteroscopy, of which ∼100,000 are performed annually in England with a 5%-10% cancer yield."
Huang J, Li Z, Chen E, Liang G, Zhao X, Lan M, et al. (2026). AI-augmented MALDI-TOF MS screening reveals a high burden of undiagnosed monoclonal gammopathy in adult patients. iScience. 10.1016/j.isci.2026.116923 — solo el resumen - no se pudo obtener el texto completo — el pasaje: "Monoclonal gammopathy of undetermined significance is a common precursor of multiple myeloma, yet its prevalence and clinical distribution remain poorly defined due to the limited scalability of conventional electrophoretic workflows."
Huang J, Li Z, Chen E, Liang G, Zhao X, Lan M, et al. (2026). AI-augmented MALDI-TOF MS screening reveals a high burden of undiagnosed monoclonal gammopathy in adult patients. iScience. 10.1016/j.isci.2026.116923 — solo el resumen - no se pudo obtener el texto completo — el pasaje: "Here, we developed an artificial intelligence-augmented matrix-assisted laser desorption/ionization time-of-flight mass spectrometry (MALDI-TOF MS) platform that integrates machine learning, rule-based detection of weak monoclonal signals and glycosylation assessment for automated M-protein screening."
Huang J, Li Z, Chen E, Liang G, Zhao X, Lan M, et al. (2026). AI-augmented MALDI-TOF MS screening reveals a high burden of undiagnosed monoclonal gammopathy in adult patients. iScience. 10.1016/j.isci.2026.116923 — solo el resumen - no se pudo obtener el texto completo — el pasaje: "Following training and validation on 5,218 retrospective serum samples, the platform was deployed in a real-world cohort of 12,263 adult patients."
Yang J, He Q, Peng J, Wang Y, Li J, Li H, et al. (2026). Artificial intelligence‐assisted histopathological diagnosis of endocervical gastric‐type adenocarcinoma: a multicenter model development and validation study. The Journal of Pathology: Clinical Research. 10.1002/2056-4538.70113 — solo el resumen - no se pudo obtener el texto completo — el pasaje: "Endocervical gastric-type adenocarcinoma (GAS) is one of the most aggressive subtypes of cervical cancer and is frequently underdiagnosed due to morphological ambiguity, leading to delayed diagnosis."
Yang J, He Q, Peng J, Wang Y, Li J, Li H, et al. (2026). Artificial intelligence‐assisted histopathological diagnosis of endocervical gastric‐type adenocarcinoma: a multicenter model development and validation study. The Journal of Pathology: Clinical Research. 10.1002/2056-4538.70113 — solo el resumen - no se pudo obtener el texto completo — el pasaje: "This study therefore proposes a highly sensitive artificial intelligence (AI)-assisted diagnostic system for GAS based exclusively on H&E-stained histopathological images."
Yang J, He Q, Peng J, Wang Y, Li J, Li H, et al. (2026). Artificial intelligence‐assisted histopathological diagnosis of endocervical gastric‐type adenocarcinoma: a multicenter model development and validation study. The Journal of Pathology: Clinical Research. 10.1002/2056-4538.70113 — solo el resumen - no se pudo obtener el texto completo — el pasaje: "In large-scale real-world deployment (Stage III, n = 7,056), GASPath achieved a balanced accuracy of 0.953, with 100% sensitivity for GAS (45/45 cases correctly identified)."
Jing L, Ye B, Liu S. (2026). Evaluation of an AI-assisted digital health follow-up system integrating humanistic care for patients undergoing chemotherapy: a prospective quasi-experimental study. Frontiers in Public Health. 10.3389/fpubh.2026.1863027 — solo el resumen - no se pudo obtener el texto completo — el pasaje: "The nurse-supervised AI-assisted follow-up system integrated rule-based risk monitoring, DeepSeek-based large language model-assisted communication, personalized follow-up, and data visualization."
Jing L, Ye B, Liu S. (2026). Evaluation of an AI-assisted digital health follow-up system integrating humanistic care for patients undergoing chemotherapy: a prospective quasi-experimental study. Frontiers in Public Health. 10.3389/fpubh.2026.1863027 — solo el resumen - no se pudo obtener el texto completo — el pasaje: "Chemotherapy imposes substantial physical and psychological burdens that may reduce treatment adherence and quality of life. Traditional follow-up is often fragmented, reactive, and insufficiently personalized."
Jing L, Ye B, Liu S. (2026). Evaluation of an AI-assisted digital health follow-up system integrating humanistic care for patients undergoing chemotherapy: a prospective quasi-experimental study. Frontiers in Public Health. 10.3389/fpubh.2026.1863027 — solo el resumen - no se pudo obtener el texto completo — el pasaje: "Randomized multicenter studies are required to confirm these findings."
Neal, M., Dean, M., Duffy, S. et al. (2026). Real-World Validation of PinPoint Blood Tests in the NHS: Multivariable Machine Learning to Predict Cancer Risk in Primary Care Urgent Referrals. Mayo Clinic Proceedings: Digital Health. https://doi.org/10.1016/j.mcpdig.2026.100382 - el artículo del que trata esta nota — el artículo completo — el pasaje: "This work shows the potential of these tests to improve urgent suspected cancer referral pathways. High-risk patients could be diagnosed more rapidly, leading to potential earlier-stage diagnosis and a better diagnostic experience. Low-risk patients could avoid unnecessary invasive medical testing for cancer. The software can be deployed rapidly across the NHS, without the need for additional hardware."

No tome esto como consejo médico profesional.

experiment · Mayo Clinic proceedings. Digital health · the paper, 10 Jun 2026 · free

A Blood Test That Sorts Cancer Referrals, Tested but Not Yet Used

In one English region, a risk score sorted 13,255 referred patients well in five cancer pathways. No one was diagnosed earlier because of it, and the study cannot say how it would work elsewhere.

Short version · the longer version follows, about 1 min

Ask the weeklyAI chat

Ask me about this study: who was studied, what it found, and what it does not say.

Conversations are saved for as long as weeklyAI exists, to improve the publication. Answers come in the language you write in.

The study at a glance
Who: people referred urgently for suspected cancer
How many: 16,481 patients
Where: West Yorkshire and Harrogate, England
When: December 2020 to July 2025
Kind of study: analysis of what people did
Who did it: PinPoint Data Science, NHS trusts, and universities in England
The limit that matters: No doctor saw the results, so nobody was diagnosed earlier or treated differently.

The results were calculated and sent to a laboratory, and no doctor saw them.
weeklyAI's reading

How it could look · illustration generated by weeklyAI.watch, not a photograph

You get an urgent referral for possible cancer. Then come the appointments, the scans, the waits. A blood test already taken in that first visit might help doctors decide who needs those tests soonest.

That is what researchers in West Yorkshire and Harrogate, in England, set out to measure. The PinPoint Tests use age, sex and the results of blood tests you would have anyway, run through a computer model, to estimate a person's cancer risk.

They followed 16,481 patients referred urgently for suspected cancer between December 2020 and July 2025, across five NHS hospital trusts and 170 family doctor surgeries. Of those, 13,255 were included in the final analysis. Cancer was found in 871 of them, about 1 in 15.

For five pathways, the score sorted patients well. The best was upper gut cancer. For upper gut cancer, if doctors had tested only the 10 percent of patients the score ranked highest, they would have needed 6.1 times fewer investigations to find one cancer. The other four were gynecological, lung, head and neck, and lower gut. Across the five tests, the reduction ranged from 2.6 to 6.1 times.

Here is what that does not mean. The results were calculated and sent to a laboratory, and no doctor saw them. Nobody was diagnosed earlier, and nobody's treatment changed. The benefit shown is a possibility, not something that happened to a patient.

The study also ran in one part of one country, inside the NHS. It cannot show how the score would behave in another health system, or for groups it could not check: ethnicity information was missing for most patients, so nothing can be said about how the test works for different ethnic groups.

If a test like this reaches where you live, it would sit alongside your doctor's judgment, not replace it. It is meant to sort referrals, not to rule cancer in or out on its own.

Watch for news that a test like this is being used in ordinary care, and that someone is counting what happens to patients because of it.

What this means for you

For now, this changes nothing about your own care, since no doctor saw the score and nobody was diagnosed earlier because of it. If a blood test like this arrives where you live, ask whether it is meant to sort referrals alongside your doctor's judgment, and watch for anyone counting what happens to patients afterward.

Who paid: PinPoint Data Science funded the contributions of several authors, and aspects of the work were supported by an award from SBRI Healthcare and the West Yorkshire and Harrogate Cancer Alliance; the article does not state whether funders had any say in the study.

Do not take this as professional medical advice.

The longer versionThe passages copied from the paper, the pictures, and every source with how much of it we read · 47 words · about 1 minRead it →Close

A Blood Test Sorted Cancer Referrals by Risk. It Was Never Allowed to Change Anyone's Care.

In an NHS region, machine learning read routine blood results from 16,481 people referred urgently for suspected cancer. It found the high-risk patients well. It was never used to treat them.

In the National Health Service in England, more than 3 million people a year are referred urgently by their family doctor on suspicion of cancer, and the referral rate has climbed 10 percent every year for fifteen years. Only about 6 percent of those referrals turn out to be cancer1. The rest go through scans, cameras and biopsies to reach a negative answer.

The PinPoint Tests are blood tests meant to sit inside that bottleneck. They are not new laboratory hardware; they are software marked for medical use in the United Kingdom, described as multi-cancer early detection tests for predicting the cancer risk of people who already have symptoms2.

The way it works is deliberately unglamorous. The model takes your age, your sex, and the blood results a laboratory already produces on a routine request: full blood count, liver function tests, urea and electrolytes, bone profile, inflammatory markers, and a set of tumor markers3. A different combination is used for each of nine referral pathways, from breast to urological. Nothing extra is drawn, nothing extra is measured. The software reads the numbers your doctor already ordered and returns a risk score.

The evaluation ran from December 2020 to July 2025 in West Yorkshire and Harrogate, across five hospital trusts and 170 general practices. It enrolled 16,481 people referred on urgent suspected cancer pathways4. It was prospective and observational: patients were followed forward in time, but the test result was not used to decide anything56.

Five of the nine tests showed performance the authors call potentially clinically useful. On the measure of how well a test separates cancer from non-cancer, upper gastrointestinal scored 0.86, gynecological 0.81, lung 0.79, head and neck 0.73, and lower gastrointestinal 0.72, each with a range of uncertainty7. If the top 10 percent of highest-risk patients were investigated first, the number of people who had to be tested to find one cancer fell by a factor of 2.6 to 6.17. Four of those five tests reached a negative predictive value above 0.99 when used to set aside the lowest-risk fifth of patients8.

Picture what the prioritization number means rather than the statistic. In a clinic where a doctor must work through a waiting list, sorting by risk means the same number of procedures finds more cancers, because the people most likely to have one are moved to the front. The consequence of that consequence is the part the article is careful about: high-risk patients could be diagnosed more rapidly and low-risk patients could avoid invasive testing, but that is stated as what could follow, not what this study watched happen9.

Here is the limit that governs everything above. The test results were calculated and sent to the regional laboratory for aggregation, and were not used clinically6. Clinicians were blinded. No patient's scan, biopsy or treatment was moved earlier or later because of this blood test. The benefit shown is a measurement of sorting power, not evidence that anyone did better.

Two other limits matter. About 3,197 people, or 19.4 percent of those enrolled, were excluded from the main analysis for various reasons, which the authors say could be a source of bias10. And referral rules themselves changed during the five years: national guidance for bowel referrals was updated in 2020, 2021 and 2023, with smaller updates to prostate referrals in 2021 and myeloma referrals in 2025, which will have affected those pathways.

The authors are not neutral observers. Eight named authors are employed by PinPoint Data Science and hold shares or options in it. The University of Leeds and Leeds Teaching Hospitals Trust hold a royalty agreement with the company, and one professor is a named inventor on it11. The company funded those authors' contributions.

This belongs to a family of stories about software reading routine medical data for hidden signals. In the upper gastrointestinal pathway alone, roughly 250,000 referrals a year arrive with a cancer yield near 3 percent and no triage tool at all; the head and neck pathway sees about 285,000 referrals a year, also without routine triage, and only a third of its cancers are currently found early1213. The gynecological test was validated mostly against endometrial cancer, and the authors suggest it could reduce demand for hysteroscopy, of which about 100,000 are performed annually in England with a 5 to 10 percent cancer yield14.

Here is how we read it. When a system has more people to investigate than it can investigate at once, it will sort them, and any score it uses becomes a quiet judgement about who is seen first and who waits. That is not a flaw in this particular test; it is what triage is. What we would expect, in homes like yours, is that a low score starts to feel like an answer rather than a step. What would show we are wrong is a clear safety net: a re-check, a route back in if symptoms continue, and a clinician who says out loud that the score does not close the question. So if a blood result ever tells you your risk is low, ask what happens next if your symptoms do not go away, and who you should call. Ask whether the test is meant to replace further checks or only to set their order.

We would add one more thing we would want to see before trusting the sorting. A warning is only useful if the door behind it opens, and only if someone has checked that it opens the same way for everyone. The people most likely to be missed by a risk score are the ones least represented in the data behind it, and this evaluation cannot answer that question at all: ethnicity information was missing for a large share of patients in most pathways, and the authors state plainly that no conclusions can be drawn about performance across ethnic groups. Ask your clinic whether a test like this has been checked for people like you, and ask what the concrete next step is if your result comes back flagged.

The authors' own next step is deployment. They write that the software can be rolled out rapidly across the NHS without additional hardware9, which is a claim about logistics, not about outcomes. Whether it improves survival, or shortens the time to a diagnosis in real clinics, is a question this study was not built to answer.

As we read it, this is what the story makes possible for you, and nothing more. The next time someone you love is referred urgently for suspected cancer, you will know that a version of this exists: a test that uses blood already being drawn, to say who should jump the queue and who might be spared an invasive procedure. You will also know the two questions that decide whether it helps your family or merely reorganizes the wait. Ask what happens if the symptoms continue after a low result. And ask whether anyone has checked that the test works the same for you.

---

Where each piece of context comes from, and how much of it we read

Neal, M., Dean, M., Duffy, S. et al. (2026). Real-World Validation of PinPoint Blood Tests in the NHS: Multivariable Machine Learning to Predict Cancer Risk in Primary Care Urgent Referrals. Mayo Clinic Proceedings: Digital Health. https://doi.org/10.1016/j.mcpdig.2026.100382 - the article this story is about — the whole article — the passage: "Urgent referrals from primary care are a major route for cancer diagnosis in the National Health service (NHS), with more than 3 million patients referred annually in England alone. The annual referral rate has increased 10% year-on-year for the last 15 years, 1 and the urgent suspected cancer (USC) pathways have an average conversion rate of only 6%, meaning that improved methods of triage are urgently needed."
Neal, M., Dean, M., Duffy, S. et al. (2026). Real-World Validation of PinPoint Blood Tests in the NHS: Multivariable Machine Learning to Predict Cancer Risk in Primary Care Urgent Referrals. Mayo Clinic Proceedings: Digital Health. https://doi.org/10.1016/j.mcpdig.2026.100382 - the article this story is about — the whole article — the passage: "The PinPoint Tests are a set of United Kingdom Conformity Assessed-marked multi-cancer early detection blood tests for predicting the cancer risk of symptomatic patients."
Neal, M., Dean, M., Duffy, S. et al. (2026). Real-World Validation of PinPoint Blood Tests in the NHS: Multivariable Machine Learning to Predict Cancer Risk in Primary Care Urgent Referrals. Mayo Clinic Proceedings: Digital Health. https://doi.org/10.1016/j.mcpdig.2026.100382 - the article this story is about — the whole article — the passage: "The predictors are age, sex, and a panel of standard blood analytes, including full blood count, liver function tests, urea and electrolytes, bone profile, inflammatory markers, and a set of tumor markers."
Neal, M., Dean, M., Duffy, S. et al. (2026). Real-World Validation of PinPoint Blood Tests in the NHS: Multivariable Machine Learning to Predict Cancer Risk in Primary Care Urgent Referrals. Mayo Clinic Proceedings: Digital Health. https://doi.org/10.1016/j.mcpdig.2026.100382 - the article this story is about — the whole article — the passage: "Total of 16,481 patients with urgent suspected cancer referrals were enrolled across 5 secondary care Trusts and 170 General Practitioner surgeries."
Neal, M., Dean, M., Duffy, S. et al. (2026). Real-World Validation of PinPoint Blood Tests in the NHS: Multivariable Machine Learning to Predict Cancer Risk in Primary Care Urgent Referrals. Mayo Clinic Proceedings: Digital Health. https://doi.org/10.1016/j.mcpdig.2026.100382 - the article this story is about — the whole article — the passage: "This work comprises a large-scale, prospective, observational, real-world NHS service evaluation of nine United Kingdom Conformity Assessed-marked Software as a Medical Device multi-cancer early detection blood tests."
Neal, M., Dean, M., Duffy, S. et al. (2026). Real-World Validation of PinPoint Blood Tests in the NHS: Multivariable Machine Learning to Predict Cancer Risk in Primary Care Urgent Referrals. Mayo Clinic Proceedings: Digital Health. https://doi.org/10.1016/j.mcpdig.2026.100382 - the article this story is about — the whole article — the passage: "The evaluation was observational. PinPoint Test results were calculated and returned to the regional hub laboratory for aggregation but were not used clinically."
Neal, M., Dean, M., Duffy, S. et al. (2026). Real-World Validation of PinPoint Blood Tests in the NHS: Multivariable Machine Learning to Predict Cancer Risk in Primary Care Urgent Referrals. Mayo Clinic Proceedings: Digital Health. https://doi.org/10.1016/j.mcpdig.2026.100382 - the article this story is about — the whole article — the passage: "Five tests have performance indicating potential clinical utility. Receiver operating characteristic area-under-curve scores (95% CI) for these were: Upper gastrointestinal=0.86 (0.81-0.90), Gynecological=0.81 (0.77-0.85), Lung=0.79 (0.74-0.84), Head & Neck=0.73 (0.68-0.78), and Lower gastrointestinal=0.72 (0.67-0.78), Prioritization of the 10% of highest-risk patients would reduce the number needed to investigate to detect one cancer by a factor of 2.6-6.1."
Neal, M., Dean, M., Duffy, S. et al. (2026). Real-World Validation of PinPoint Blood Tests in the NHS: Multivariable Machine Learning to Predict Cancer Risk in Primary Care Urgent Referrals. Mayo Clinic Proceedings: Digital Health. https://doi.org/10.1016/j.mcpdig.2026.100382 - the article this story is about — the whole article — the passage: "Four of these five tests achieved an negative predictive value >0.99 when used to rule-out 20% of the lowest-risk patients."
Neal, M., Dean, M., Duffy, S. et al. (2026). Real-World Validation of PinPoint Blood Tests in the NHS: Multivariable Machine Learning to Predict Cancer Risk in Primary Care Urgent Referrals. Mayo Clinic Proceedings: Digital Health. https://doi.org/10.1016/j.mcpdig.2026.100382 - the article this story is about — the whole article — the passage: "This work shows the potential of these tests to improve urgent suspected cancer referral pathways. High-risk patients could be diagnosed more rapidly, leading to potential earlier-stage diagnosis and a better diagnostic experience. Low-risk patients could avoid unnecessary invasive medical testing for cancer. The software can be deployed rapidly across the NHS, without the need for additional hardware."
Neal, M., Dean, M., Duffy, S. et al. (2026). Real-World Validation of PinPoint Blood Tests in the NHS: Multivariable Machine Learning to Predict Cancer Risk in Primary Care Urgent Referrals. Mayo Clinic Proceedings: Digital Health. https://doi.org/10.1016/j.mcpdig.2026.100382 - the article this story is about — the whole article — the passage: "About 3,197 (19.4%) patients were excluded from the analysis for various reasons."
Neal, M., Dean, M., Duffy, S. et al. (2026). Real-World Validation of PinPoint Blood Tests in the NHS: Multivariable Machine Learning to Predict Cancer Risk in Primary Care Urgent Referrals. Mayo Clinic Proceedings: Digital Health. https://doi.org/10.1016/j.mcpdig.2026.100382 - the article this story is about — the whole article — the passage: "Drs Savage, Lloyd, Neal, Skinner, Sansom, Tully, Ferguson, and Duffy are employed by, and are shareholders or option holders in, PinPoint Data Science. Both the University of Leeds and Leeds Teaching Hospitals Trust have a royalty agreement with PinPoint Data Science; Prof Richard Neal is a named inventor in this royalty agreement."
Neal, M., Dean, M., Duffy, S. et al. (2026). Real-World Validation of PinPoint Blood Tests in the NHS: Multivariable Machine Learning to Predict Cancer Risk in Primary Care Urgent Referrals. Mayo Clinic Proceedings: Digital Health. https://doi.org/10.1016/j.mcpdig.2026.100382 - the article this story is about — the whole article — the passage: "The Upper GI test, which has the strongest performance, addresses a high-volume (250,000 referrals/year), low-yield (∼3%) pathway with no current triage tool."
Neal, M., Dean, M., Duffy, S. et al. (2026). Real-World Validation of PinPoint Blood Tests in the NHS: Multivariable Machine Learning to Predict Cancer Risk in Primary Care Urgent Referrals. Mayo Clinic Proceedings: Digital Health. https://doi.org/10.1016/j.mcpdig.2026.100382 - the article this story is about — the whole article — the passage: "The Head & Neck test addresses another high-volume (285,000 referrals/year) pathway with no routine triage tool, where only one third of cancers are currently diagnosed early and only half start treatment within the NHS 62-day target."
Neal, M., Dean, M., Duffy, S. et al. (2026). Real-World Validation of PinPoint Blood Tests in the NHS: Multivariable Machine Learning to Predict Cancer Risk in Primary Care Urgent Referrals. Mayo Clinic Proceedings: Digital Health. https://doi.org/10.1016/j.mcpdig.2026.100382 - the article this story is about — the whole article — the passage: "The gynecological test (validated primarily against endometrial cancer, 61% of cohort cancers) could reduce demand for hysteroscopy, of which ∼100,000 are performed annually in England with a 5%-10% cancer yield."

Do not take this as professional medical advice.